Parameter prediction for RIU-LBP based on PSO-BP algorithm
Ying Tan, Yuchun Fang, Cheng Ying Gong, Dai Wang · 2011
Local Binary Pattern (LBP) is one of the most popular feature extraction algorithms in face recognition with good performance. However, setting proper parameters for this algorithm is still an open question in pattern recognition. In most previous research, this problem is solved with experienced comparison tests. However, such tests might be constrained by certain database and application and thus lack of generalization ability. In this paper, based on our previous research on factor analysis of the Rotation Invariant Uniform LBP (RIU-LBP) feature, we propose a parameter prediction and selection method based on the Particle Swarm Optimizer-Back Propagation neural network (PSO-BP) for setting the dominant factor, i.e. the blocking number for RIU-LBP feature. Experimental results show that the proposed prediction method could effectively save the computation time in parameter selection.